Toyota Research Institute Unveils Breakthrough in Teaching Robots New Behaviors

20 September 2023

Today, the Toyota Research Institute (TRI) announced a significant breakthrough in the field of robotics with its new generative AI approach based on Diffusion Policy. This approach is designed to efficiently and confidently teach robots new dexterous skills, revolutionizing their utility and capabilities.

Gill Pratt, CEO of TRI and Chief Scientist for Toyota Motor Corporation, emphasized that the aim of their robotics research is to amplify human abilities rather than replace them. The new teaching technique is highly efficient and produces high-performing behaviors in robots, enabling them to effectively assist humans in various ways.

Previous methods for teaching robots new behaviors were slow, inconsistent, and often limited to specific tasks in controlled environments. Roboticists would spend significant time writing complex code or conducting trial-and-error experiments to program behaviors.

TRI has already successfully taught robots more than 60 challenging dexterous skills using their new approach, including tasks like pouring liquids, using tools, and manipulating deformable objects. Notably, this was achieved without writing any new code; the robots simply needed new data. TRI's ambitious goal is to teach hundreds of additional skills by the end of the year and 1,000 by the end of 2024.

This breakthrough demonstrates that robots can be taught to function in diverse scenarios and perform a wide range of behaviors, extending beyond simple "pick and place" tasks. TRI's robots are becoming capable of interacting with the world in rich and varied ways, making them valuable assets in supporting people in everyday and unpredictable environments.